]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame_incremental - arch/x86/kernel/cpu/common.c
x86/cpufeature: Cleanup get_cpu_cap()
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kernel / cpu / common.c
... / ...
CommitLineData
1#include <linux/bootmem.h>
2#include <linux/linkage.h>
3#include <linux/bitops.h>
4#include <linux/kernel.h>
5#include <linux/module.h>
6#include <linux/percpu.h>
7#include <linux/string.h>
8#include <linux/ctype.h>
9#include <linux/delay.h>
10#include <linux/sched.h>
11#include <linux/init.h>
12#include <linux/kprobes.h>
13#include <linux/kgdb.h>
14#include <linux/smp.h>
15#include <linux/io.h>
16#include <linux/syscore_ops.h>
17
18#include <asm/stackprotector.h>
19#include <asm/perf_event.h>
20#include <asm/mmu_context.h>
21#include <asm/archrandom.h>
22#include <asm/hypervisor.h>
23#include <asm/processor.h>
24#include <asm/tlbflush.h>
25#include <asm/debugreg.h>
26#include <asm/sections.h>
27#include <asm/vsyscall.h>
28#include <linux/topology.h>
29#include <linux/cpumask.h>
30#include <asm/pgtable.h>
31#include <linux/atomic.h>
32#include <asm/proto.h>
33#include <asm/setup.h>
34#include <asm/apic.h>
35#include <asm/desc.h>
36#include <asm/fpu/internal.h>
37#include <asm/mtrr.h>
38#include <linux/numa.h>
39#include <asm/asm.h>
40#include <asm/cpu.h>
41#include <asm/mce.h>
42#include <asm/msr.h>
43#include <asm/pat.h>
44#include <asm/microcode.h>
45#include <asm/microcode_intel.h>
46
47#ifdef CONFIG_X86_LOCAL_APIC
48#include <asm/uv/uv.h>
49#endif
50
51#include "cpu.h"
52
53/* all of these masks are initialized in setup_cpu_local_masks() */
54cpumask_var_t cpu_initialized_mask;
55cpumask_var_t cpu_callout_mask;
56cpumask_var_t cpu_callin_mask;
57
58/* representing cpus for which sibling maps can be computed */
59cpumask_var_t cpu_sibling_setup_mask;
60
61/* correctly size the local cpu masks */
62void __init setup_cpu_local_masks(void)
63{
64 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
65 alloc_bootmem_cpumask_var(&cpu_callin_mask);
66 alloc_bootmem_cpumask_var(&cpu_callout_mask);
67 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
68}
69
70static void default_init(struct cpuinfo_x86 *c)
71{
72#ifdef CONFIG_X86_64
73 cpu_detect_cache_sizes(c);
74#else
75 /* Not much we can do here... */
76 /* Check if at least it has cpuid */
77 if (c->cpuid_level == -1) {
78 /* No cpuid. It must be an ancient CPU */
79 if (c->x86 == 4)
80 strcpy(c->x86_model_id, "486");
81 else if (c->x86 == 3)
82 strcpy(c->x86_model_id, "386");
83 }
84#endif
85}
86
87static const struct cpu_dev default_cpu = {
88 .c_init = default_init,
89 .c_vendor = "Unknown",
90 .c_x86_vendor = X86_VENDOR_UNKNOWN,
91};
92
93static const struct cpu_dev *this_cpu = &default_cpu;
94
95DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
96#ifdef CONFIG_X86_64
97 /*
98 * We need valid kernel segments for data and code in long mode too
99 * IRET will check the segment types kkeil 2000/10/28
100 * Also sysret mandates a special GDT layout
101 *
102 * TLS descriptors are currently at a different place compared to i386.
103 * Hopefully nobody expects them at a fixed place (Wine?)
104 */
105 [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
106 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
107 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
108 [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
109 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
110 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
111#else
112 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
113 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
114 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
115 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
116 /*
117 * Segments used for calling PnP BIOS have byte granularity.
118 * They code segments and data segments have fixed 64k limits,
119 * the transfer segment sizes are set at run time.
120 */
121 /* 32-bit code */
122 [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
123 /* 16-bit code */
124 [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
125 /* 16-bit data */
126 [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
127 /* 16-bit data */
128 [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0),
129 /* 16-bit data */
130 [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0),
131 /*
132 * The APM segments have byte granularity and their bases
133 * are set at run time. All have 64k limits.
134 */
135 /* 32-bit code */
136 [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
137 /* 16-bit code */
138 [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
139 /* data */
140 [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
141
142 [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
143 [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
144 GDT_STACK_CANARY_INIT
145#endif
146} };
147EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
148
149static int __init x86_mpx_setup(char *s)
150{
151 /* require an exact match without trailing characters */
152 if (strlen(s))
153 return 0;
154
155 /* do not emit a message if the feature is not present */
156 if (!boot_cpu_has(X86_FEATURE_MPX))
157 return 1;
158
159 setup_clear_cpu_cap(X86_FEATURE_MPX);
160 pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
161 return 1;
162}
163__setup("nompx", x86_mpx_setup);
164
165#ifdef CONFIG_X86_32
166static int cachesize_override = -1;
167static int disable_x86_serial_nr = 1;
168
169static int __init cachesize_setup(char *str)
170{
171 get_option(&str, &cachesize_override);
172 return 1;
173}
174__setup("cachesize=", cachesize_setup);
175
176static int __init x86_sep_setup(char *s)
177{
178 setup_clear_cpu_cap(X86_FEATURE_SEP);
179 return 1;
180}
181__setup("nosep", x86_sep_setup);
182
183/* Standard macro to see if a specific flag is changeable */
184static inline int flag_is_changeable_p(u32 flag)
185{
186 u32 f1, f2;
187
188 /*
189 * Cyrix and IDT cpus allow disabling of CPUID
190 * so the code below may return different results
191 * when it is executed before and after enabling
192 * the CPUID. Add "volatile" to not allow gcc to
193 * optimize the subsequent calls to this function.
194 */
195 asm volatile ("pushfl \n\t"
196 "pushfl \n\t"
197 "popl %0 \n\t"
198 "movl %0, %1 \n\t"
199 "xorl %2, %0 \n\t"
200 "pushl %0 \n\t"
201 "popfl \n\t"
202 "pushfl \n\t"
203 "popl %0 \n\t"
204 "popfl \n\t"
205
206 : "=&r" (f1), "=&r" (f2)
207 : "ir" (flag));
208
209 return ((f1^f2) & flag) != 0;
210}
211
212/* Probe for the CPUID instruction */
213int have_cpuid_p(void)
214{
215 return flag_is_changeable_p(X86_EFLAGS_ID);
216}
217
218static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
219{
220 unsigned long lo, hi;
221
222 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
223 return;
224
225 /* Disable processor serial number: */
226
227 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
228 lo |= 0x200000;
229 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
230
231 printk(KERN_NOTICE "CPU serial number disabled.\n");
232 clear_cpu_cap(c, X86_FEATURE_PN);
233
234 /* Disabling the serial number may affect the cpuid level */
235 c->cpuid_level = cpuid_eax(0);
236}
237
238static int __init x86_serial_nr_setup(char *s)
239{
240 disable_x86_serial_nr = 0;
241 return 1;
242}
243__setup("serialnumber", x86_serial_nr_setup);
244#else
245static inline int flag_is_changeable_p(u32 flag)
246{
247 return 1;
248}
249static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
250{
251}
252#endif
253
254static __init int setup_disable_smep(char *arg)
255{
256 setup_clear_cpu_cap(X86_FEATURE_SMEP);
257 return 1;
258}
259__setup("nosmep", setup_disable_smep);
260
261static __always_inline void setup_smep(struct cpuinfo_x86 *c)
262{
263 if (cpu_has(c, X86_FEATURE_SMEP))
264 cr4_set_bits(X86_CR4_SMEP);
265}
266
267static __init int setup_disable_smap(char *arg)
268{
269 setup_clear_cpu_cap(X86_FEATURE_SMAP);
270 return 1;
271}
272__setup("nosmap", setup_disable_smap);
273
274static __always_inline void setup_smap(struct cpuinfo_x86 *c)
275{
276 unsigned long eflags = native_save_fl();
277
278 /* This should have been cleared long ago */
279 BUG_ON(eflags & X86_EFLAGS_AC);
280
281 if (cpu_has(c, X86_FEATURE_SMAP)) {
282#ifdef CONFIG_X86_SMAP
283 cr4_set_bits(X86_CR4_SMAP);
284#else
285 cr4_clear_bits(X86_CR4_SMAP);
286#endif
287 }
288}
289
290/*
291 * Some CPU features depend on higher CPUID levels, which may not always
292 * be available due to CPUID level capping or broken virtualization
293 * software. Add those features to this table to auto-disable them.
294 */
295struct cpuid_dependent_feature {
296 u32 feature;
297 u32 level;
298};
299
300static const struct cpuid_dependent_feature
301cpuid_dependent_features[] = {
302 { X86_FEATURE_MWAIT, 0x00000005 },
303 { X86_FEATURE_DCA, 0x00000009 },
304 { X86_FEATURE_XSAVE, 0x0000000d },
305 { 0, 0 }
306};
307
308static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
309{
310 const struct cpuid_dependent_feature *df;
311
312 for (df = cpuid_dependent_features; df->feature; df++) {
313
314 if (!cpu_has(c, df->feature))
315 continue;
316 /*
317 * Note: cpuid_level is set to -1 if unavailable, but
318 * extended_extended_level is set to 0 if unavailable
319 * and the legitimate extended levels are all negative
320 * when signed; hence the weird messing around with
321 * signs here...
322 */
323 if (!((s32)df->level < 0 ?
324 (u32)df->level > (u32)c->extended_cpuid_level :
325 (s32)df->level > (s32)c->cpuid_level))
326 continue;
327
328 clear_cpu_cap(c, df->feature);
329 if (!warn)
330 continue;
331
332 printk(KERN_WARNING
333 "CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
334 x86_cap_flag(df->feature), df->level);
335 }
336}
337
338/*
339 * Naming convention should be: <Name> [(<Codename>)]
340 * This table only is used unless init_<vendor>() below doesn't set it;
341 * in particular, if CPUID levels 0x80000002..4 are supported, this
342 * isn't used
343 */
344
345/* Look up CPU names by table lookup. */
346static const char *table_lookup_model(struct cpuinfo_x86 *c)
347{
348#ifdef CONFIG_X86_32
349 const struct legacy_cpu_model_info *info;
350
351 if (c->x86_model >= 16)
352 return NULL; /* Range check */
353
354 if (!this_cpu)
355 return NULL;
356
357 info = this_cpu->legacy_models;
358
359 while (info->family) {
360 if (info->family == c->x86)
361 return info->model_names[c->x86_model];
362 info++;
363 }
364#endif
365 return NULL; /* Not found */
366}
367
368__u32 cpu_caps_cleared[NCAPINTS];
369__u32 cpu_caps_set[NCAPINTS];
370
371void load_percpu_segment(int cpu)
372{
373#ifdef CONFIG_X86_32
374 loadsegment(fs, __KERNEL_PERCPU);
375#else
376 loadsegment(gs, 0);
377 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
378#endif
379 load_stack_canary_segment();
380}
381
382/*
383 * Current gdt points %fs at the "master" per-cpu area: after this,
384 * it's on the real one.
385 */
386void switch_to_new_gdt(int cpu)
387{
388 struct desc_ptr gdt_descr;
389
390 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
391 gdt_descr.size = GDT_SIZE - 1;
392 load_gdt(&gdt_descr);
393 /* Reload the per-cpu base */
394
395 load_percpu_segment(cpu);
396}
397
398static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
399
400static void get_model_name(struct cpuinfo_x86 *c)
401{
402 unsigned int *v;
403 char *p, *q, *s;
404
405 if (c->extended_cpuid_level < 0x80000004)
406 return;
407
408 v = (unsigned int *)c->x86_model_id;
409 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
410 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
411 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
412 c->x86_model_id[48] = 0;
413
414 /* Trim whitespace */
415 p = q = s = &c->x86_model_id[0];
416
417 while (*p == ' ')
418 p++;
419
420 while (*p) {
421 /* Note the last non-whitespace index */
422 if (!isspace(*p))
423 s = q;
424
425 *q++ = *p++;
426 }
427
428 *(s + 1) = '\0';
429}
430
431void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
432{
433 unsigned int n, dummy, ebx, ecx, edx, l2size;
434
435 n = c->extended_cpuid_level;
436
437 if (n >= 0x80000005) {
438 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
439 c->x86_cache_size = (ecx>>24) + (edx>>24);
440#ifdef CONFIG_X86_64
441 /* On K8 L1 TLB is inclusive, so don't count it */
442 c->x86_tlbsize = 0;
443#endif
444 }
445
446 if (n < 0x80000006) /* Some chips just has a large L1. */
447 return;
448
449 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
450 l2size = ecx >> 16;
451
452#ifdef CONFIG_X86_64
453 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
454#else
455 /* do processor-specific cache resizing */
456 if (this_cpu->legacy_cache_size)
457 l2size = this_cpu->legacy_cache_size(c, l2size);
458
459 /* Allow user to override all this if necessary. */
460 if (cachesize_override != -1)
461 l2size = cachesize_override;
462
463 if (l2size == 0)
464 return; /* Again, no L2 cache is possible */
465#endif
466
467 c->x86_cache_size = l2size;
468}
469
470u16 __read_mostly tlb_lli_4k[NR_INFO];
471u16 __read_mostly tlb_lli_2m[NR_INFO];
472u16 __read_mostly tlb_lli_4m[NR_INFO];
473u16 __read_mostly tlb_lld_4k[NR_INFO];
474u16 __read_mostly tlb_lld_2m[NR_INFO];
475u16 __read_mostly tlb_lld_4m[NR_INFO];
476u16 __read_mostly tlb_lld_1g[NR_INFO];
477
478static void cpu_detect_tlb(struct cpuinfo_x86 *c)
479{
480 if (this_cpu->c_detect_tlb)
481 this_cpu->c_detect_tlb(c);
482
483 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
484 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
485 tlb_lli_4m[ENTRIES]);
486
487 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
488 tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
489 tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
490}
491
492void detect_ht(struct cpuinfo_x86 *c)
493{
494#ifdef CONFIG_SMP
495 u32 eax, ebx, ecx, edx;
496 int index_msb, core_bits;
497 static bool printed;
498
499 if (!cpu_has(c, X86_FEATURE_HT))
500 return;
501
502 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
503 goto out;
504
505 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
506 return;
507
508 cpuid(1, &eax, &ebx, &ecx, &edx);
509
510 smp_num_siblings = (ebx & 0xff0000) >> 16;
511
512 if (smp_num_siblings == 1) {
513 printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
514 goto out;
515 }
516
517 if (smp_num_siblings <= 1)
518 goto out;
519
520 index_msb = get_count_order(smp_num_siblings);
521 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
522
523 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
524
525 index_msb = get_count_order(smp_num_siblings);
526
527 core_bits = get_count_order(c->x86_max_cores);
528
529 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
530 ((1 << core_bits) - 1);
531
532out:
533 if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
534 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
535 c->phys_proc_id);
536 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
537 c->cpu_core_id);
538 printed = 1;
539 }
540#endif
541}
542
543static void get_cpu_vendor(struct cpuinfo_x86 *c)
544{
545 char *v = c->x86_vendor_id;
546 int i;
547
548 for (i = 0; i < X86_VENDOR_NUM; i++) {
549 if (!cpu_devs[i])
550 break;
551
552 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
553 (cpu_devs[i]->c_ident[1] &&
554 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
555
556 this_cpu = cpu_devs[i];
557 c->x86_vendor = this_cpu->c_x86_vendor;
558 return;
559 }
560 }
561
562 printk_once(KERN_ERR
563 "CPU: vendor_id '%s' unknown, using generic init.\n" \
564 "CPU: Your system may be unstable.\n", v);
565
566 c->x86_vendor = X86_VENDOR_UNKNOWN;
567 this_cpu = &default_cpu;
568}
569
570void cpu_detect(struct cpuinfo_x86 *c)
571{
572 /* Get vendor name */
573 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
574 (unsigned int *)&c->x86_vendor_id[0],
575 (unsigned int *)&c->x86_vendor_id[8],
576 (unsigned int *)&c->x86_vendor_id[4]);
577
578 c->x86 = 4;
579 /* Intel-defined flags: level 0x00000001 */
580 if (c->cpuid_level >= 0x00000001) {
581 u32 junk, tfms, cap0, misc;
582
583 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
584 c->x86 = (tfms >> 8) & 0xf;
585 c->x86_model = (tfms >> 4) & 0xf;
586 c->x86_mask = tfms & 0xf;
587
588 if (c->x86 == 0xf)
589 c->x86 += (tfms >> 20) & 0xff;
590 if (c->x86 >= 0x6)
591 c->x86_model += ((tfms >> 16) & 0xf) << 4;
592
593 if (cap0 & (1<<19)) {
594 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
595 c->x86_cache_alignment = c->x86_clflush_size;
596 }
597 }
598}
599
600void get_cpu_cap(struct cpuinfo_x86 *c)
601{
602 u32 eax, ebx, ecx, edx;
603
604 /* Intel-defined flags: level 0x00000001 */
605 if (c->cpuid_level >= 0x00000001) {
606 cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
607
608 c->x86_capability[CPUID_1_ECX] = ecx;
609 c->x86_capability[CPUID_1_EDX] = edx;
610 }
611
612 /* Additional Intel-defined flags: level 0x00000007 */
613 if (c->cpuid_level >= 0x00000007) {
614 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
615
616 c->x86_capability[CPUID_7_0_EBX] = ebx;
617
618 c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
619 }
620
621 /* Extended state features: level 0x0000000d */
622 if (c->cpuid_level >= 0x0000000d) {
623 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
624
625 c->x86_capability[CPUID_D_1_EAX] = eax;
626 }
627
628 /* Additional Intel-defined flags: level 0x0000000F */
629 if (c->cpuid_level >= 0x0000000F) {
630
631 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
632 cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
633 c->x86_capability[CPUID_F_0_EDX] = edx;
634
635 if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
636 /* will be overridden if occupancy monitoring exists */
637 c->x86_cache_max_rmid = ebx;
638
639 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
640 cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
641 c->x86_capability[CPUID_F_1_EDX] = edx;
642
643 if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
644 c->x86_cache_max_rmid = ecx;
645 c->x86_cache_occ_scale = ebx;
646 }
647 } else {
648 c->x86_cache_max_rmid = -1;
649 c->x86_cache_occ_scale = -1;
650 }
651 }
652
653 /* AMD-defined flags: level 0x80000001 */
654 eax = cpuid_eax(0x80000000);
655 c->extended_cpuid_level = eax;
656
657 if ((eax & 0xffff0000) == 0x80000000) {
658 if (eax >= 0x80000001) {
659 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
660
661 c->x86_capability[CPUID_8000_0001_ECX] = ecx;
662 c->x86_capability[CPUID_8000_0001_EDX] = edx;
663 }
664 }
665
666 if (c->extended_cpuid_level >= 0x80000008) {
667 cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
668
669 c->x86_virt_bits = (eax >> 8) & 0xff;
670 c->x86_phys_bits = eax & 0xff;
671 c->x86_capability[CPUID_8000_0008_EBX] = ebx;
672 }
673#ifdef CONFIG_X86_32
674 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
675 c->x86_phys_bits = 36;
676#endif
677
678 if (c->extended_cpuid_level >= 0x80000007)
679 c->x86_power = cpuid_edx(0x80000007);
680
681 if (c->extended_cpuid_level >= 0x8000000a)
682 c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
683
684 init_scattered_cpuid_features(c);
685}
686
687static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
688{
689#ifdef CONFIG_X86_32
690 int i;
691
692 /*
693 * First of all, decide if this is a 486 or higher
694 * It's a 486 if we can modify the AC flag
695 */
696 if (flag_is_changeable_p(X86_EFLAGS_AC))
697 c->x86 = 4;
698 else
699 c->x86 = 3;
700
701 for (i = 0; i < X86_VENDOR_NUM; i++)
702 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
703 c->x86_vendor_id[0] = 0;
704 cpu_devs[i]->c_identify(c);
705 if (c->x86_vendor_id[0]) {
706 get_cpu_vendor(c);
707 break;
708 }
709 }
710#endif
711}
712
713/*
714 * Do minimum CPU detection early.
715 * Fields really needed: vendor, cpuid_level, family, model, mask,
716 * cache alignment.
717 * The others are not touched to avoid unwanted side effects.
718 *
719 * WARNING: this function is only called on the BP. Don't add code here
720 * that is supposed to run on all CPUs.
721 */
722static void __init early_identify_cpu(struct cpuinfo_x86 *c)
723{
724#ifdef CONFIG_X86_64
725 c->x86_clflush_size = 64;
726 c->x86_phys_bits = 36;
727 c->x86_virt_bits = 48;
728#else
729 c->x86_clflush_size = 32;
730 c->x86_phys_bits = 32;
731 c->x86_virt_bits = 32;
732#endif
733 c->x86_cache_alignment = c->x86_clflush_size;
734
735 memset(&c->x86_capability, 0, sizeof c->x86_capability);
736 c->extended_cpuid_level = 0;
737
738 if (!have_cpuid_p())
739 identify_cpu_without_cpuid(c);
740
741 /* cyrix could have cpuid enabled via c_identify()*/
742 if (!have_cpuid_p())
743 return;
744
745 cpu_detect(c);
746 get_cpu_vendor(c);
747 get_cpu_cap(c);
748
749 if (this_cpu->c_early_init)
750 this_cpu->c_early_init(c);
751
752 c->cpu_index = 0;
753 filter_cpuid_features(c, false);
754
755 if (this_cpu->c_bsp_init)
756 this_cpu->c_bsp_init(c);
757
758 setup_force_cpu_cap(X86_FEATURE_ALWAYS);
759 fpu__init_system(c);
760}
761
762void __init early_cpu_init(void)
763{
764 const struct cpu_dev *const *cdev;
765 int count = 0;
766
767#ifdef CONFIG_PROCESSOR_SELECT
768 printk(KERN_INFO "KERNEL supported cpus:\n");
769#endif
770
771 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
772 const struct cpu_dev *cpudev = *cdev;
773
774 if (count >= X86_VENDOR_NUM)
775 break;
776 cpu_devs[count] = cpudev;
777 count++;
778
779#ifdef CONFIG_PROCESSOR_SELECT
780 {
781 unsigned int j;
782
783 for (j = 0; j < 2; j++) {
784 if (!cpudev->c_ident[j])
785 continue;
786 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
787 cpudev->c_ident[j]);
788 }
789 }
790#endif
791 }
792 early_identify_cpu(&boot_cpu_data);
793}
794
795/*
796 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
797 * unfortunately, that's not true in practice because of early VIA
798 * chips and (more importantly) broken virtualizers that are not easy
799 * to detect. In the latter case it doesn't even *fail* reliably, so
800 * probing for it doesn't even work. Disable it completely on 32-bit
801 * unless we can find a reliable way to detect all the broken cases.
802 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
803 */
804static void detect_nopl(struct cpuinfo_x86 *c)
805{
806#ifdef CONFIG_X86_32
807 clear_cpu_cap(c, X86_FEATURE_NOPL);
808#else
809 set_cpu_cap(c, X86_FEATURE_NOPL);
810#endif
811}
812
813static void generic_identify(struct cpuinfo_x86 *c)
814{
815 c->extended_cpuid_level = 0;
816
817 if (!have_cpuid_p())
818 identify_cpu_without_cpuid(c);
819
820 /* cyrix could have cpuid enabled via c_identify()*/
821 if (!have_cpuid_p())
822 return;
823
824 cpu_detect(c);
825
826 get_cpu_vendor(c);
827
828 get_cpu_cap(c);
829
830 if (c->cpuid_level >= 0x00000001) {
831 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
832#ifdef CONFIG_X86_32
833# ifdef CONFIG_SMP
834 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
835# else
836 c->apicid = c->initial_apicid;
837# endif
838#endif
839 c->phys_proc_id = c->initial_apicid;
840 }
841
842 get_model_name(c); /* Default name */
843
844 detect_nopl(c);
845}
846
847static void x86_init_cache_qos(struct cpuinfo_x86 *c)
848{
849 /*
850 * The heavy lifting of max_rmid and cache_occ_scale are handled
851 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
852 * in case CQM bits really aren't there in this CPU.
853 */
854 if (c != &boot_cpu_data) {
855 boot_cpu_data.x86_cache_max_rmid =
856 min(boot_cpu_data.x86_cache_max_rmid,
857 c->x86_cache_max_rmid);
858 }
859}
860
861/*
862 * This does the hard work of actually picking apart the CPU stuff...
863 */
864static void identify_cpu(struct cpuinfo_x86 *c)
865{
866 int i;
867
868 c->loops_per_jiffy = loops_per_jiffy;
869 c->x86_cache_size = -1;
870 c->x86_vendor = X86_VENDOR_UNKNOWN;
871 c->x86_model = c->x86_mask = 0; /* So far unknown... */
872 c->x86_vendor_id[0] = '\0'; /* Unset */
873 c->x86_model_id[0] = '\0'; /* Unset */
874 c->x86_max_cores = 1;
875 c->x86_coreid_bits = 0;
876#ifdef CONFIG_X86_64
877 c->x86_clflush_size = 64;
878 c->x86_phys_bits = 36;
879 c->x86_virt_bits = 48;
880#else
881 c->cpuid_level = -1; /* CPUID not detected */
882 c->x86_clflush_size = 32;
883 c->x86_phys_bits = 32;
884 c->x86_virt_bits = 32;
885#endif
886 c->x86_cache_alignment = c->x86_clflush_size;
887 memset(&c->x86_capability, 0, sizeof c->x86_capability);
888
889 generic_identify(c);
890
891 if (this_cpu->c_identify)
892 this_cpu->c_identify(c);
893
894 /* Clear/Set all flags overriden by options, after probe */
895 for (i = 0; i < NCAPINTS; i++) {
896 c->x86_capability[i] &= ~cpu_caps_cleared[i];
897 c->x86_capability[i] |= cpu_caps_set[i];
898 }
899
900#ifdef CONFIG_X86_64
901 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
902#endif
903
904 /*
905 * Vendor-specific initialization. In this section we
906 * canonicalize the feature flags, meaning if there are
907 * features a certain CPU supports which CPUID doesn't
908 * tell us, CPUID claiming incorrect flags, or other bugs,
909 * we handle them here.
910 *
911 * At the end of this section, c->x86_capability better
912 * indicate the features this CPU genuinely supports!
913 */
914 if (this_cpu->c_init)
915 this_cpu->c_init(c);
916
917 /* Disable the PN if appropriate */
918 squash_the_stupid_serial_number(c);
919
920 /* Set up SMEP/SMAP */
921 setup_smep(c);
922 setup_smap(c);
923
924 /*
925 * The vendor-specific functions might have changed features.
926 * Now we do "generic changes."
927 */
928
929 /* Filter out anything that depends on CPUID levels we don't have */
930 filter_cpuid_features(c, true);
931
932 /* If the model name is still unset, do table lookup. */
933 if (!c->x86_model_id[0]) {
934 const char *p;
935 p = table_lookup_model(c);
936 if (p)
937 strcpy(c->x86_model_id, p);
938 else
939 /* Last resort... */
940 sprintf(c->x86_model_id, "%02x/%02x",
941 c->x86, c->x86_model);
942 }
943
944#ifdef CONFIG_X86_64
945 detect_ht(c);
946#endif
947
948 init_hypervisor(c);
949 x86_init_rdrand(c);
950 x86_init_cache_qos(c);
951
952 /*
953 * Clear/Set all flags overriden by options, need do it
954 * before following smp all cpus cap AND.
955 */
956 for (i = 0; i < NCAPINTS; i++) {
957 c->x86_capability[i] &= ~cpu_caps_cleared[i];
958 c->x86_capability[i] |= cpu_caps_set[i];
959 }
960
961 /*
962 * On SMP, boot_cpu_data holds the common feature set between
963 * all CPUs; so make sure that we indicate which features are
964 * common between the CPUs. The first time this routine gets
965 * executed, c == &boot_cpu_data.
966 */
967 if (c != &boot_cpu_data) {
968 /* AND the already accumulated flags with these */
969 for (i = 0; i < NCAPINTS; i++)
970 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
971
972 /* OR, i.e. replicate the bug flags */
973 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
974 c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
975 }
976
977 /* Init Machine Check Exception if available. */
978 mcheck_cpu_init(c);
979
980 select_idle_routine(c);
981
982#ifdef CONFIG_NUMA
983 numa_add_cpu(smp_processor_id());
984#endif
985}
986
987/*
988 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
989 * on 32-bit kernels:
990 */
991#ifdef CONFIG_X86_32
992void enable_sep_cpu(void)
993{
994 struct tss_struct *tss;
995 int cpu;
996
997 cpu = get_cpu();
998 tss = &per_cpu(cpu_tss, cpu);
999
1000 if (!boot_cpu_has(X86_FEATURE_SEP))
1001 goto out;
1002
1003 /*
1004 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1005 * see the big comment in struct x86_hw_tss's definition.
1006 */
1007
1008 tss->x86_tss.ss1 = __KERNEL_CS;
1009 wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1010
1011 wrmsr(MSR_IA32_SYSENTER_ESP,
1012 (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1013 0);
1014
1015 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1016
1017out:
1018 put_cpu();
1019}
1020#endif
1021
1022void __init identify_boot_cpu(void)
1023{
1024 identify_cpu(&boot_cpu_data);
1025 init_amd_e400_c1e_mask();
1026#ifdef CONFIG_X86_32
1027 sysenter_setup();
1028 enable_sep_cpu();
1029#endif
1030 cpu_detect_tlb(&boot_cpu_data);
1031}
1032
1033void identify_secondary_cpu(struct cpuinfo_x86 *c)
1034{
1035 BUG_ON(c == &boot_cpu_data);
1036 identify_cpu(c);
1037#ifdef CONFIG_X86_32
1038 enable_sep_cpu();
1039#endif
1040 mtrr_ap_init();
1041}
1042
1043struct msr_range {
1044 unsigned min;
1045 unsigned max;
1046};
1047
1048static const struct msr_range msr_range_array[] = {
1049 { 0x00000000, 0x00000418},
1050 { 0xc0000000, 0xc000040b},
1051 { 0xc0010000, 0xc0010142},
1052 { 0xc0011000, 0xc001103b},
1053};
1054
1055static void __print_cpu_msr(void)
1056{
1057 unsigned index_min, index_max;
1058 unsigned index;
1059 u64 val;
1060 int i;
1061
1062 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1063 index_min = msr_range_array[i].min;
1064 index_max = msr_range_array[i].max;
1065
1066 for (index = index_min; index < index_max; index++) {
1067 if (rdmsrl_safe(index, &val))
1068 continue;
1069 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1070 }
1071 }
1072}
1073
1074static int show_msr;
1075
1076static __init int setup_show_msr(char *arg)
1077{
1078 int num;
1079
1080 get_option(&arg, &num);
1081
1082 if (num > 0)
1083 show_msr = num;
1084 return 1;
1085}
1086__setup("show_msr=", setup_show_msr);
1087
1088static __init int setup_noclflush(char *arg)
1089{
1090 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1091 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1092 return 1;
1093}
1094__setup("noclflush", setup_noclflush);
1095
1096void print_cpu_info(struct cpuinfo_x86 *c)
1097{
1098 const char *vendor = NULL;
1099
1100 if (c->x86_vendor < X86_VENDOR_NUM) {
1101 vendor = this_cpu->c_vendor;
1102 } else {
1103 if (c->cpuid_level >= 0)
1104 vendor = c->x86_vendor_id;
1105 }
1106
1107 if (vendor && !strstr(c->x86_model_id, vendor))
1108 printk(KERN_CONT "%s ", vendor);
1109
1110 if (c->x86_model_id[0])
1111 printk(KERN_CONT "%s", c->x86_model_id);
1112 else
1113 printk(KERN_CONT "%d86", c->x86);
1114
1115 printk(KERN_CONT " (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1116
1117 if (c->x86_mask || c->cpuid_level >= 0)
1118 printk(KERN_CONT ", stepping: 0x%x)\n", c->x86_mask);
1119 else
1120 printk(KERN_CONT ")\n");
1121
1122 print_cpu_msr(c);
1123}
1124
1125void print_cpu_msr(struct cpuinfo_x86 *c)
1126{
1127 if (c->cpu_index < show_msr)
1128 __print_cpu_msr();
1129}
1130
1131static __init int setup_disablecpuid(char *arg)
1132{
1133 int bit;
1134
1135 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1136 setup_clear_cpu_cap(bit);
1137 else
1138 return 0;
1139
1140 return 1;
1141}
1142__setup("clearcpuid=", setup_disablecpuid);
1143
1144#ifdef CONFIG_X86_64
1145struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1146struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1147 (unsigned long) debug_idt_table };
1148
1149DEFINE_PER_CPU_FIRST(union irq_stack_union,
1150 irq_stack_union) __aligned(PAGE_SIZE) __visible;
1151
1152/*
1153 * The following percpu variables are hot. Align current_task to
1154 * cacheline size such that they fall in the same cacheline.
1155 */
1156DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1157 &init_task;
1158EXPORT_PER_CPU_SYMBOL(current_task);
1159
1160DEFINE_PER_CPU(char *, irq_stack_ptr) =
1161 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1162
1163DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1164
1165DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1166EXPORT_PER_CPU_SYMBOL(__preempt_count);
1167
1168/*
1169 * Special IST stacks which the CPU switches to when it calls
1170 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1171 * limit), all of them are 4K, except the debug stack which
1172 * is 8K.
1173 */
1174static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1175 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1176 [DEBUG_STACK - 1] = DEBUG_STKSZ
1177};
1178
1179static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1180 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1181
1182/* May not be marked __init: used by software suspend */
1183void syscall_init(void)
1184{
1185 /*
1186 * LSTAR and STAR live in a bit strange symbiosis.
1187 * They both write to the same internal register. STAR allows to
1188 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1189 */
1190 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1191 wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1192
1193#ifdef CONFIG_IA32_EMULATION
1194 wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1195 /*
1196 * This only works on Intel CPUs.
1197 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1198 * This does not cause SYSENTER to jump to the wrong location, because
1199 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1200 */
1201 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1202 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1203 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1204#else
1205 wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1206 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1207 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1208 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1209#endif
1210
1211 /* Flags to clear on syscall */
1212 wrmsrl(MSR_SYSCALL_MASK,
1213 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1214 X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1215}
1216
1217/*
1218 * Copies of the original ist values from the tss are only accessed during
1219 * debugging, no special alignment required.
1220 */
1221DEFINE_PER_CPU(struct orig_ist, orig_ist);
1222
1223static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1224DEFINE_PER_CPU(int, debug_stack_usage);
1225
1226int is_debug_stack(unsigned long addr)
1227{
1228 return __this_cpu_read(debug_stack_usage) ||
1229 (addr <= __this_cpu_read(debug_stack_addr) &&
1230 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1231}
1232NOKPROBE_SYMBOL(is_debug_stack);
1233
1234DEFINE_PER_CPU(u32, debug_idt_ctr);
1235
1236void debug_stack_set_zero(void)
1237{
1238 this_cpu_inc(debug_idt_ctr);
1239 load_current_idt();
1240}
1241NOKPROBE_SYMBOL(debug_stack_set_zero);
1242
1243void debug_stack_reset(void)
1244{
1245 if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1246 return;
1247 if (this_cpu_dec_return(debug_idt_ctr) == 0)
1248 load_current_idt();
1249}
1250NOKPROBE_SYMBOL(debug_stack_reset);
1251
1252#else /* CONFIG_X86_64 */
1253
1254DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1255EXPORT_PER_CPU_SYMBOL(current_task);
1256DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1257EXPORT_PER_CPU_SYMBOL(__preempt_count);
1258
1259/*
1260 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1261 * the top of the kernel stack. Use an extra percpu variable to track the
1262 * top of the kernel stack directly.
1263 */
1264DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1265 (unsigned long)&init_thread_union + THREAD_SIZE;
1266EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1267
1268#ifdef CONFIG_CC_STACKPROTECTOR
1269DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1270#endif
1271
1272#endif /* CONFIG_X86_64 */
1273
1274/*
1275 * Clear all 6 debug registers:
1276 */
1277static void clear_all_debug_regs(void)
1278{
1279 int i;
1280
1281 for (i = 0; i < 8; i++) {
1282 /* Ignore db4, db5 */
1283 if ((i == 4) || (i == 5))
1284 continue;
1285
1286 set_debugreg(0, i);
1287 }
1288}
1289
1290#ifdef CONFIG_KGDB
1291/*
1292 * Restore debug regs if using kgdbwait and you have a kernel debugger
1293 * connection established.
1294 */
1295static void dbg_restore_debug_regs(void)
1296{
1297 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1298 arch_kgdb_ops.correct_hw_break();
1299}
1300#else /* ! CONFIG_KGDB */
1301#define dbg_restore_debug_regs()
1302#endif /* ! CONFIG_KGDB */
1303
1304static void wait_for_master_cpu(int cpu)
1305{
1306#ifdef CONFIG_SMP
1307 /*
1308 * wait for ACK from master CPU before continuing
1309 * with AP initialization
1310 */
1311 WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1312 while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1313 cpu_relax();
1314#endif
1315}
1316
1317/*
1318 * cpu_init() initializes state that is per-CPU. Some data is already
1319 * initialized (naturally) in the bootstrap process, such as the GDT
1320 * and IDT. We reload them nevertheless, this function acts as a
1321 * 'CPU state barrier', nothing should get across.
1322 * A lot of state is already set up in PDA init for 64 bit
1323 */
1324#ifdef CONFIG_X86_64
1325
1326void cpu_init(void)
1327{
1328 struct orig_ist *oist;
1329 struct task_struct *me;
1330 struct tss_struct *t;
1331 unsigned long v;
1332 int cpu = stack_smp_processor_id();
1333 int i;
1334
1335 wait_for_master_cpu(cpu);
1336
1337 /*
1338 * Initialize the CR4 shadow before doing anything that could
1339 * try to read it.
1340 */
1341 cr4_init_shadow();
1342
1343 /*
1344 * Load microcode on this cpu if a valid microcode is available.
1345 * This is early microcode loading procedure.
1346 */
1347 load_ucode_ap();
1348
1349 t = &per_cpu(cpu_tss, cpu);
1350 oist = &per_cpu(orig_ist, cpu);
1351
1352#ifdef CONFIG_NUMA
1353 if (this_cpu_read(numa_node) == 0 &&
1354 early_cpu_to_node(cpu) != NUMA_NO_NODE)
1355 set_numa_node(early_cpu_to_node(cpu));
1356#endif
1357
1358 me = current;
1359
1360 pr_debug("Initializing CPU#%d\n", cpu);
1361
1362 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1363
1364 /*
1365 * Initialize the per-CPU GDT with the boot GDT,
1366 * and set up the GDT descriptor:
1367 */
1368
1369 switch_to_new_gdt(cpu);
1370 loadsegment(fs, 0);
1371
1372 load_current_idt();
1373
1374 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1375 syscall_init();
1376
1377 wrmsrl(MSR_FS_BASE, 0);
1378 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1379 barrier();
1380
1381 x86_configure_nx();
1382 x2apic_setup();
1383
1384 /*
1385 * set up and load the per-CPU TSS
1386 */
1387 if (!oist->ist[0]) {
1388 char *estacks = per_cpu(exception_stacks, cpu);
1389
1390 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1391 estacks += exception_stack_sizes[v];
1392 oist->ist[v] = t->x86_tss.ist[v] =
1393 (unsigned long)estacks;
1394 if (v == DEBUG_STACK-1)
1395 per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1396 }
1397 }
1398
1399 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1400
1401 /*
1402 * <= is required because the CPU will access up to
1403 * 8 bits beyond the end of the IO permission bitmap.
1404 */
1405 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1406 t->io_bitmap[i] = ~0UL;
1407
1408 atomic_inc(&init_mm.mm_count);
1409 me->active_mm = &init_mm;
1410 BUG_ON(me->mm);
1411 enter_lazy_tlb(&init_mm, me);
1412
1413 load_sp0(t, &current->thread);
1414 set_tss_desc(cpu, t);
1415 load_TR_desc();
1416 load_mm_ldt(&init_mm);
1417
1418 clear_all_debug_regs();
1419 dbg_restore_debug_regs();
1420
1421 fpu__init_cpu();
1422
1423 if (is_uv_system())
1424 uv_cpu_init();
1425}
1426
1427#else
1428
1429void cpu_init(void)
1430{
1431 int cpu = smp_processor_id();
1432 struct task_struct *curr = current;
1433 struct tss_struct *t = &per_cpu(cpu_tss, cpu);
1434 struct thread_struct *thread = &curr->thread;
1435
1436 wait_for_master_cpu(cpu);
1437
1438 /*
1439 * Initialize the CR4 shadow before doing anything that could
1440 * try to read it.
1441 */
1442 cr4_init_shadow();
1443
1444 show_ucode_info_early();
1445
1446 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1447
1448 if (cpu_feature_enabled(X86_FEATURE_VME) || cpu_has_tsc || cpu_has_de)
1449 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1450
1451 load_current_idt();
1452 switch_to_new_gdt(cpu);
1453
1454 /*
1455 * Set up and load the per-CPU TSS and LDT
1456 */
1457 atomic_inc(&init_mm.mm_count);
1458 curr->active_mm = &init_mm;
1459 BUG_ON(curr->mm);
1460 enter_lazy_tlb(&init_mm, curr);
1461
1462 load_sp0(t, thread);
1463 set_tss_desc(cpu, t);
1464 load_TR_desc();
1465 load_mm_ldt(&init_mm);
1466
1467 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1468
1469#ifdef CONFIG_DOUBLEFAULT
1470 /* Set up doublefault TSS pointer in the GDT */
1471 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1472#endif
1473
1474 clear_all_debug_regs();
1475 dbg_restore_debug_regs();
1476
1477 fpu__init_cpu();
1478}
1479#endif
1480
1481#ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1482void warn_pre_alternatives(void)
1483{
1484 WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1485}
1486EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1487#endif
1488
1489inline bool __static_cpu_has_safe(u16 bit)
1490{
1491 return boot_cpu_has(bit);
1492}
1493EXPORT_SYMBOL_GPL(__static_cpu_has_safe);
1494
1495static void bsp_resume(void)
1496{
1497 if (this_cpu->c_bsp_resume)
1498 this_cpu->c_bsp_resume(&boot_cpu_data);
1499}
1500
1501static struct syscore_ops cpu_syscore_ops = {
1502 .resume = bsp_resume,
1503};
1504
1505static int __init init_cpu_syscore(void)
1506{
1507 register_syscore_ops(&cpu_syscore_ops);
1508 return 0;
1509}
1510core_initcall(init_cpu_syscore);